[LoopVectorize] Improve algorithm for hoisting runtime checks (#73515)

When attempting to hoist runtime checks out of a loop we currently avoid
creating pointer diff checks and prefer to do expanded range checks
instead. This gives us the opportunity to hoist runtime checks out of a
loop, since these checks are loop invariant. However, in some cases the
pointer diff checks would also be loop invariant and so will naturally
get hoisted. Therefore, since diff checks are cheaper so we should
prefer to use those instead.

Author: david-arm

llvm/lib/Analysis/LoopAccessAnalysis.cpp

@@ -347,7 +347,12 @@ void RuntimePointerChecking::tryToCreateDiffCheck(
     auto *SinkStartAR = cast<SCEVAddRecExpr>(SinkStartInt);
     const Loop *StartARLoop = SrcStartAR->getLoop();
     if (StartARLoop == SinkStartAR->getLoop() &&
-        StartARLoop == InnerLoop->getParentLoop()) {
+        StartARLoop == InnerLoop->getParentLoop() &&
+        // If the diff check would already be loop invariant (due to the
+        // recurrences being the same), then we prefer to keep the diff checks
+        // because they are cheaper.
+        SrcStartAR->getStepRecurrence(*SE) !=
+            SinkStartAR->getStepRecurrence(*SE)) {
       LLVM_DEBUG(dbgs() << "LAA: Not creating diff runtime check, since these "
                            "cannot be hoisted out of the outer loop\n");
       CanUseDiffCheck = false;